Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
1. A method performed by one or more computer systems coupled to a packet-based network to predict locations of mobile devices, each of the one or more computer systems including at least one processor, the method comprising: determining, by one or more processors of the one or more computer systems, location events associated with a plurality of mobile devices communicating with the packet-based network, each location event having a time stamp and indicating an associated mobile device being at one or more of a plurality of geo-places; machine training a plurality of off-line prediction models using at least location events in a first time period and location events in a first plurality of time frames after the first time period, the plurality of off-line prediction models corresponding to respective ones of the first plurality of time frames; generating, by one or more processors of the one or more computer systems, off-line prediction results for one or more mobile devices by applying the plurality of off-line prediction models to one or more sets of features corresponding, respectively, to the one or more mobile devices, the one or more sets of features including features derived from location events associated with the one or more mobile devices in a second time period, the second time period having a start time after a start time of the first time period, the off-line prediction results including predicted probabilities of a particular mobile device of the one or more mobile devices being at one or more pre-selected locations during respective ones of a second plurality of time frames after the second time period; machine training an on-line prediction model using at least location events in a third time period, location events in the first plurality of time frames, and location events in a third time frame after the third time period; generating, by one or more processors of the one or more computer systems, an on-line prediction result associated with the particular mobile device by applying the on-line prediction model to a set of features for the particular mobile device, the set of features including features related to the predicted probabilities, and features derived from location events associated with the particular mobile device in a fourth time period, the fourth time period being shorter than the first or second time period, the on-line prediction results indicating a probability of the particular mobile device being at any of the one or more pre-selected locations during a fourth time frame after the fourth time period; and in response to receiving from the packet-based network a request associated with the particular mobile device, processing the request based at least on the on-line prediction result associated with the particular mobile device.
This invention relates to predicting the locations of mobile devices using machine learning models. The system addresses the challenge of accurately forecasting mobile device locations over time, which is useful for applications like network optimization, targeted advertising, or emergency services. The method involves collecting location events from multiple mobile devices, where each event includes a timestamp and indicates the device's presence at specific geo-places. These events are used to train multiple off-line prediction models, each corresponding to different time frames. The models are trained using historical location data from a first time period and subsequent time frames. Features derived from location events in a second time period are then applied to these models to generate off-line prediction results, which include probabilities of a device being at pre-selected locations in future time frames. Additionally, an on-line prediction model is trained using location data from a third time period, the initial time frames, and an additional time frame. This model is applied to features derived from recent location events (within a shorter fourth time period) and the off-line prediction results to generate an on-line prediction result. This result indicates the probability of a device being at any of the pre-selected locations in a future time frame. When a request related to a specific mobile device is received, the system processes it based on the on-line prediction result, enabling real-time decision-making. The approach combines historical and recent data to improve prediction accuracy.
2. The method of claim 1 , wherein: the third time frame is shorter than a shortest one of the first plurality of time frames; and the fourth time frame is shorter than a shortest one of the second plurality of time frames.
3. The method of claim 1 , wherein machine training a plurality of off-line prediction models comprises: constructing a first feature space for the plurality of mobile device, including generating features using location events associated with the plurality of mobile devices in the first time period; extracting a plurality of sets of labels related to the one or more pre-selected locations from location events associated with the plurality of mobile devices in the first plurality of time frames; and machine training the plurality of off-line prediction models using the first feature space and respective sets of the plurality of sets of labels.
This invention relates to predictive modeling for mobile devices, specifically improving location-based predictions by training multiple off-line models using historical location data. The problem addressed is the need for accurate and efficient prediction of mobile device behavior, particularly in relation to pre-selected locations, by leveraging large-scale historical data. The method involves constructing a feature space for a plurality of mobile devices by generating features from location events recorded during a first time period. These features capture patterns in device movement and interactions with specific locations. Additionally, labels are extracted from location events associated with the same devices during multiple time frames, where these labels correspond to the pre-selected locations of interest. The labels represent whether a device was present at or interacted with these locations during the observed time frames. Using the constructed feature space and the extracted labels, multiple off-line prediction models are trained. These models are designed to predict future device behavior, such as the likelihood of a device visiting a pre-selected location. The training process ensures that the models can generalize from historical data to make accurate predictions. The use of multiple models allows for robustness and flexibility in handling different types of location-based predictions. This approach enhances the reliability of location-based services and applications by improving the accuracy of predictions based on historical mobility patterns.
4. The method of claim 3 , wherein the first feature space includes a set of features corresponding to each of the plurality of mobile devices.
A system and method for analyzing mobile device behavior involves monitoring and processing data from multiple mobile devices to detect anomalies or security threats. The method includes generating a first feature space that represents a set of features for each mobile device, where these features are derived from device activity, network interactions, or other behavioral data. This feature space is used to identify patterns, deviations, or malicious activities by comparing individual device features against a baseline or predefined thresholds. The system may also incorporate a second feature space that aggregates features across multiple devices to detect coordinated or widespread threats. By analyzing these feature spaces, the system can identify compromised devices, unauthorized access attempts, or other security risks in real time. The method may further include applying machine learning models to classify or predict threats based on the extracted features. This approach enhances mobile security by providing a comprehensive view of device behavior and enabling proactive threat detection.
5. The method of claim 3 , wherein machine training an on-line prediction model comprises: constructing a second feature space for the plurality of mobile devices, the second feature space including features related to the plurality of sets of labels, and features derived from location events associated with the plurality of mobile devices in the third time period; extracting, by one or more processors of the one or more computer systems, an additional set of labels related to the one or more pre-selected locations from location events associated with the plurality of mobile devices in the third time frame; and training the on-line prediction model using the second feature space, and the additional set of labels.
This invention relates to training machine learning models for predicting mobile device behavior based on location data. The problem addressed is improving the accuracy of predictions by leveraging historical and real-time location events to refine model training. The method involves constructing a feature space that includes both labeled data and derived features from location events of multiple mobile devices over a specific time period. The feature space is used to train an online prediction model, which is continuously updated to enhance its predictive capabilities. The training process extracts additional labels from location events associated with pre-selected locations, further refining the model's ability to predict device behavior. The system processes these features and labels using one or more computer systems to ensure the model adapts to new data dynamically. This approach aims to improve the reliability of predictions by incorporating diverse and up-to-date location-based information.
6. The method of claim 5 , wherein the plurality of geo-places include a plurality of geo-fences, each of the plurality of geo-fences corresponding to a point of interest of a plurality of points of interest (POIs), and wherein the second feature space further includes at least some of: features related to a plurality of brands, each of the plurality of brands being associated with at least one of the plurality of POIs; and features related to most-recently visited brands.
This invention relates to location-based data analysis, specifically methods for processing and analyzing geo-location data to extract meaningful insights. The problem addressed involves efficiently categorizing and interpreting geo-location data, particularly in the context of points of interest (POIs) and user behavior patterns. The method involves defining a plurality of geo-places, which include geo-fences corresponding to specific POIs. Each geo-fence represents a bounded area around a POI, enabling precise tracking of user interactions with that location. The system processes this data to generate a second feature space, which includes features derived from user interactions with brands associated with these POIs. The feature space incorporates brand-related data, where each brand is linked to one or more POIs, and also includes features based on the most-recently visited brands by users. This allows for the analysis of user preferences, brand engagement, and behavioral trends over time. By integrating geo-fence data with brand and visit history, the method enables more accurate profiling of user behavior, which can be used for targeted marketing, personalized recommendations, or location-based services. The approach enhances the granularity of location data analysis by correlating physical visits with brand interactions, providing deeper insights into consumer habits.
7. The method of claim 6 , wherein the plurality of geo-places further include a plurality of geo-blocks, each of the plurality of geo-blocks corresponding to a geographical region having at least one border defined by a public road or natural boundary, and wherein the second feature space further includes at least some of: features related to most-recently-triggered geo-blocks among the plurality of geo-blocks; features related to most-frequently-visited geo-blocks among the plurality of geo-blocks; features related to one or more retail geo-blocks among the plurality of geo-blocks; a feature for each of the plurality of mobile devices and related to a number of distinct POIs at which the each mobile device is located during the third time period; a feature for each of the plurality of mobile devices and related to a number of distinct geo-blocks in which the each mobile device is located during the first time period; and a feature for each of the plurality of mobile devices and related to a number of visits made by a user of the each mobile device to any geo-blocks among the plurality of geo-blocks during the first time period.
8. The method of claim 7 , further comprising: dividing the plurality of geo-blocks into a number of geo-block brackets each corresponding to a distinct range of relevance measures with respect to the one or more pre-selected locations; and wherein the second feature space further includes features related, respectively, to the number of geo-block brackets.
9. The method of claim 8 , wherein the second feature space further includes at least some of: features related to non-location data associated with the plurality of mobile devices; features related to impression events related to one or more documents associated with the one or more pre-selected locations during the third time period; and features related to click/call events related to the one or more documents associated with the one or more pre-selected locations during the third time period.
This invention relates to analyzing mobile device data to evaluate the effectiveness of location-based advertising. The system identifies a set of pre-selected locations and collects data from mobile devices that were present at those locations during a specified time period. The collected data includes location-based features, such as device dwell time and movement patterns, as well as non-location data like device characteristics or user behavior. The system also tracks impression and click/call events associated with documents (e.g., ads) linked to the pre-selected locations during the same time period. These features are used to generate a second feature space, which is combined with a first feature space derived from baseline location data. The combined feature space is then analyzed to assess the impact of the advertising campaign, such as determining whether the ads influenced user behavior or engagement. The method enables advertisers to measure the effectiveness of location-based marketing by correlating mobile device activity with ad exposure and subsequent actions.
10. The method of claim 1 , wherein the set of features for the particular mobile device further includes features related to at least some of: a day of a week when the on-line prediction result is generated; a time of the day when the on-line prediction result is generated; a speed of travel of the particular mobile device near the time; a road on which the particular mobile device is traveling; a city in which the particular mobile device is located near the time; a most-recently used mobile app used on the particular mobile device; and non-location data associated with the particular mobile device.
11. A method performed by one or more computer systems coupled to a packet-based network to predict mobile device locations, each of the one or more computer systems including at least one processor, the method comprising: determining, by one or more processors of the one or more computer systems, location events associated with mobile devices communicating with the packet-based network, each location event having a time stamp and identifying one or more of a plurality of geo-places, the plurality of geo-places including a plurality of geo-fences and a plurality of geo-blocks, the plurality of geo-fences including geo-fences corresponding to a plurality of points of interest (POIs), the plurality of geo-blocks including definitions of geographical regions bordered on at least one side by public roads or natural boundaries; constructing, by one or more processors of the one or more computer systems, a first feature space for a first plurality of mobile devices, the first feature space including features derived from location events associated with the first plurality of mobile devices in a first time period; extracting, by one or more processors of the one or more computer systems, a first set of labels related to one or more pre-selected locations from location events associated with the first plurality of mobile devices in a first time frame after the first time period; machine training a first off-line location prediction model using the first feature space and the first set of labels; generating, by one or more processors of the one or more computer systems, one or more first prediction results for one or more mobile devices by applying the first off-line prediction model to one or more sets of features corresponding, respectively, to the one or more mobile devices, the one or more sets of features including features derived from location events associated with the one or more mobile devices in a second time period having an end time after an end of the first time period, the one or more first prediction results including predicted probabilities of the one or more mobile devices being at the one or more pre-selected locations during a second of time frame after the second time period; extracting, from location events associated with the first plurality of mobile devices in one or more third time frames after the first time period, one or more second sets of labels related to the one or more pre-selected locations and corresponding, respectively, to the one or more third time frames; machine training an on-line prediction model using at least location events in a third time period, location events in the first time frame and the one or more third time frames, and location events in a fourth time frame after the third time period; receiving a request associated with a particular mobile device of the one or more mobile device from the packet-based network; applying the on-line prediction model to a feature set for the particular mobile devices constructed using at least location events associated with the particular mobile device during a fifth time period to generate a second prediction result including a probability of the particular mobile device being at the one or more pre-selected locations during a fifth time frame after the fifth time period; and processing the request based at least on a predicted probability of the particular mobile device being at the one or more pre-selected locations.
12. The method of claim 11 , wherein the first feature space includes at least some of: features related to a plurality of brands, each of the plurality of brands being associated with at least one of the plurality of POIs; and features related to most-recently visited brands.
13. The method of claim 12 , wherein the first feature space further includes at least some of: features related to most-recently-triggered geo-blocks among the plurality of geo-blocks; features related to most-frequently-visited geo-blocks among the plurality of geo-blocks; features related to one or more retail geo-blocks among the plurality of geo-blocks; a feature for each of the first plurality of mobile devices and related to a number of distinct POIs at which the each mobile device is located during the first time period; a feature for each of the first plurality of mobile devices and related to a number of distinct geo-blocks in which the each mobile device is located during the first time period; and a feature for each of the first plurality of mobile devices and related to a number of visits made by a user of the each mobile device to any geo-blocks among the plurality of geo-blocks during the first time period.
14. The method of claim 11 , further comprising: dividing the plurality of geo-blocks into a number of geo-block brackets each corresponding to a distinct range of relevance measures with respect to the one or more pre-selected locations; wherein the first feature space includes features related, respectively, to the number of geo-block brackets.
15. The method of claim 11 , wherein the first feature space further includes at least some of: features related to non-location data associated with the first plurality of mobile devices; features related to impression events related to one or more documents associated with the one or more pre-selected locations during the first time period; and features related to click/call events related to the one or more documents associated with the one or more pre-selected locations during the first time period.
16. The method of claim 11 , wherein the fourth time frame is shorter than a shortest one of the first time frame and the one or more third time frames.
17. The method of claim 11 , further comprising: generating second prediction results for the one or more mobile devices by applying the one or more second prediction models to the one or more sets of features, the second prediction results including predicted probabilities of the one or more mobile devices being at the one or more pre-selected locations during one or more sixth time frames after the second time period.
18. The method of claim 17 , wherein: the feature set for the particular mobile devices is further constructed using predicted probabilities for the particular mobile device during the one or more sixth time frames after the second time period.
19. The method of claim 18 , wherein the fifth time frame is shorter than a shortest one of the second time frame and the one or more sixth time frames.
20. One or more non-transitory computer media accessible by one or more processors in one or more computer systems coupled to a packet-based network, the one or more non-transitory computer media storing therein computer readable instructions, which, when executed by the one or more processors, cause the one or more processors to perform a method comprising: machine training a plurality of off-line prediction models using at least location events in a first time period and location events in a first plurality of time frames after the first time period, the plurality of off-line prediction models corresponding to respective ones of the first plurality of time frames, each location event being associated with one of a plurality of mobile devices communicating with the packet-based network, and indicating an associated mobile device being at one or more of a plurality of geo-places at a respective time; generating off-line prediction results for one or more mobile devices by applying the plurality of off-line prediction models to one or more sets of features corresponding, respectively, to the one or more mobile devices, the one or more sets of features including features derived from location events associated with the one or more mobile devices in a second time period, the second time period having a start time after a start time of the first time period, the off-line prediction results including predicted probabilities of a particular mobile device of the one or more mobile devices being at one or more pre-selected locations during respective ones of a second plurality of time frames after the second time period; machine training an on-line prediction model using at least location events in a third time period, location events in the first plurality of time frames, and location events in a third time frame after the third time period; generating an on-line prediction result associated with the particular mobile device by applying the on-line prediction model to a set of features for the particular mobile device, the set of features including features related to the predicted probabilities of the particular mobile device, and features derived from location events associated with the particular mobile device in a fourth time period, the fourth time period being shorter than the first or second time period, the on-line prediction results indicating a probability of the particular mobile device being at any of the one or more pre-selected locations during a fourth time frame after the fourth time period; and in response to receiving from the packet-based network a request associated with the particular mobile device, processing the request based at least on a predicted probability associated with the particular mobile device.
Unknown
March 2, 2021
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.